Linguistic and psycholinguistic foundations

CHAPTER 6


Linguistic and psycholinguistic foundations


Josée Poirier and Lewis P. Shapiro


As speakers (or signers), we select words that match what we want to say, combine them appropriately, and hope our message gets through. As comprehenders, we are presented with a complex string of sounds, signs or letters and must reconstruct their associated meaning. Yet, every day, we speak/write and hear/read sentences that have never been produced before. Moreover, producing or comprehending language is fast and relatively effortless (at least in one’s native language), but language itself is tremendously complex. How do we manage the intricate knowledge we have about our language and smoothly use it to communicate?


These are core issues in psycholinguistics, the study of the psychological underpinnings of language. In this chapter we first introduce linguistic concepts that describe the complexity of language structure. Crucially, these concepts allow us to study and then detail the mental operations involved in language processing (normal and disordered). With this toolkit under our belts, we next turn to sentence processing and sentence comprehension in particular. We examine sentence comprehension from different viewpoints to explore various ways language could be represented and processed. Our aim is for the reader to acquire descriptive tools and to gain exposure to a variety of cognitive processes so that a critical and rigorous study of language processing can be perused.



Linguistics toolkit


The goal of this section is to describe the features (or properties) of verbs, with an eye toward sentence processing. Why verbs? Because:



Before we begin our tutorial about verbs, there are some important preliminaries. We begin with the following assumption: The sentence is the basic unit of analysis in language processing. Our intuitions tell us this must be so. After all, we seem to speak in sentences (embedded in discourse), and we are all amazed when young children begin to sequence words into what appear to be sentence-like units. But there are also some very simple facts that suggest that this assumption must be correct. For example, consider the following:



These three sequences of words seem to express the same proposition (kissed: John, Mary), yet the order of the words is distinct in each case. Moreover, in the 1950s (considered to be the time when the “modern era” in linguistics and psycholinguistics began) George Miller and his colleagues conducted several experiments that showed that the accurate perception of words in noise significantly increases when the words are strung together into sentences. Thus, at the very least, the sentence holds a privileged role in perception and production, and it will turn out, in comprehension as well.1


Yet it is obvious that words, too, play a critical role. Here, we assume that humans are equipped with a mental dictionary or Lexicon. The Lexicon, in an abstract sense, is the depository of knowledge about words. This knowledge includes, at least, lexical category information (part-of-speech), phonology (the sound structure of the word), meaning, and grammatical constraints. As a brief example of the latter, consider the verb kiss again. We can say, “John kissed Mary” (sentence (1), above), but if we say, “John kissed,” it somehow feels incomplete. Yet it is perfectly OK to say, “John slept.” These simple facts suggest that there are restrictions or constraints on what types of sentences in which a verb can be inserted. The verb kiss seems to require two objects (here, John and Mary) expressed in the sentence, while the verb sleep is perfectly happy acting with just a single object (John). Thus verbs are selective; they choose their syntactic and semantic partners. Our view of the lexicon, then, is that its primary role is to support well-formed sentences. That is, our ability to use words to refer to the world is not independent of the sentences in which those words are contained. This fact suggests that investigations of word-level processing, including research into word-finding difficulties and treatment of such difficulties, must eventually make contact with sentence processing. Given the primary role of the sentence, we now continue with some syntax, and how words are put together into sentences.



Merge and phrase structure


Words are not just linked up sequentially to form sentences; instead, sentences are formed from the hierarchical ordering of words. To see what we mean, consider a headline from the news:



A brief consideration of this sentence reveals an ambiguity: It can mean that some very tall doctors have sued hospitals; alternatively and more likely, it can mean that seven Podiatrists sued the hospitals. This ambiguity is structural, as can be seen in the following:



Using labeled bracketing to describe the natural divisions (i.e., constituents) in the sentences, in (5a) the subject noun phrase (NP) is structured such that [Seven foot] forms a constituent and so does [doctors], and the combination of the two yields a complex NP, Seven foot doctors. In (5b), [Seven] forms a single constituent, and when combined with [foot doctors] yields the complex NP that serves as the subject of the sentence. Note that this ambiguity does not arise from any lexical ambiguity (as in, for example, “Child’s stool is great for use in garden!”). Instead, the ambiguity arises because of two possible structures, with each structure governing a particular interpretation.


Another way of describing or viewing this structural ambiguity is through the use of hierarchical tree structures. Consider again (5a) and (5b) but represented graphically as depicted in Figure 6-1.



In (a), the word Seven forms a single constituent, while foot and doctors join to form a higher order constituent (at the intersection of the two branches), and when these are joined together we have Seven “foot doctors.” In (b), the words Seven and foot combine to form a constituent, which then joins doctors to form the entire phrase, and hence we have Seven foot doctors. These representations are the tree structure analogues of the bracketing found in (5a) and (5b). The take-home message here is that there is no way to describe this ambiguity without making reference to the structure of the phrase, and that structure is hierarchical and not simply a linear string of words.


In fact, what we have done in Figure 6-1 is compute an approximation of phrase structure, which describes the phrasal geometry of sentences. These are node-labeled tree structures with hierarchical ordering. There are lexical nodes, which refer to lexical categories (i.e., parts-of-speech), like N (Noun), V (Verb), and P (Preposition). These lexical categories form the heads of the higher-order phrases to which they project.2 So, the Verb is the head of the Verb Phrase (VP), the Preposition the head of the Prepositional Phrase (PP), and so on. Consider Figure 6-2.



The syntactic component of the grammar includes an operation, MERGE, which takes two categories as input and then outputs a single, merged, category. So, in Figure 6-2, the Determiner and Noun merge to form a higher-order NP; the Verb and NP merge to form a higher-order VP, and the VP and subject NP merge to form the Sentence.


Continuing, and keeping with the focus of this section, we now consider only the VP. VPs expand to include several possibilities; Figure 6-3 shows only three. In (a), the verb sleep has no complements; that is, it has nothing that comes after it. In (b), the verb kiss takes an NP complement. Thus, the V merges with an NP and yields the VP (or alternatively, we can say that the VP expands to include an NP complement). Finally, the verb say merges with a complement phrase/sentential clause (S) to form the VP. In this case, the embedded S would, itself, expand (as shown in Figure 6-2). So, each verb selects a particular syntactic structure. In this way, we have shown that verbs directly influence the syntax of the sentences in which they are contained.



To generate a sentence, we begin by enabling a set of lexical items (technically, a numeration) and then use successive merger operations. For example, beginning with the numeration: [Kiss; V; girl, boy], we would merge the Verb kiss with its NP (boy) to form the VP (see (8b)). We then select girl, which is an NP, and merge this with the previously formed VP, to yield the Sentence Node (see Figure 6-2 for more details). Note that the successive merger operations are not intended to mimic real-time processing of sentences (if it did, we would be parsing sentences backward!); instead, Merge is considered a linguistic operation. It remains for empirical work to discover if Merge has psycholinguistic consequences (and indeed, as we shall show shortly, there is such evidence).


We now consider another example, with the verb think [thinks, V; girl, boy] substituting for the verb kiss. So, in Figure 6-4, we go through successive merger operations (i.e., DET merges with N to form an NP, V merges with NP to form VP, NP and VP merge to form an S) to derive the sentence: “The girl thinks the boy.” Of course, our intuitions strongly suggest a distinction between the output of Merge in (see Figure 6-3, B) relative to the output in (see Figure 6-4); the former is well-formed or grammatical (“the girl kisses the boy”), while the latter is ill-formed or ungrammatical (*“the girl thinks the boy,” with * signifying a sentence is ungrammatical). Merge seems to be too powerful as we have used it here—it generates both grammatical and ungrammatical sentences. Thus we need a way to restrict the output of Merge to include only well-formed sentences. The solution turns out to involve verb properties; these will allow the theory to form only grammatical sentences. We turn to this solution in subsequent sections.




Argument structure


Consider the following verbs and the sentences in which they are contained:



Each verb needs partners to describe an event or activity. The verb disappear requires one participant as shown in (6a); the verb kiss, shown in Figure 6-4 as well as in (7a), needs two participants; and the verb put requires three as shown in (8a). To see that this is true, consider the (b) versions above, which are all ungrammatical. That is, disappear cannot occur in a sentence with two participants; kiss cannot appear in a sentence with only one participant; and put cannot appear in a sentence with two participants (or even one, as in *the girl put). Thus, verbs, again, are said to select their sentence environments.


One way to describe the restrictions on a verb’s environments is through predicate-argument structure. Borrowing from logic, we can say that sentences are composed of a verb (i.e., predicate) and a set of arguments. A verb denotes an activity or event, and an argument denotes a participant in the event. Thus, the verb disappear is a one-place predicate because it selects for a single argument (played by the NP the boy in (6a)). The verb kiss is a two-place predicate because it selects two arguments, both a subject NP and an object NP (7); and the verb put requires three arguments (8a). The maximum number of arguments that can appear with any given verb seems to be three. The minimum number is one, though there is a class of verbs that appear to express an event without any arguments; that is, they can stand on their own. This class is weather verbs (e.g., It is raining, It is snowing, etc.). Here, the subject (It, a pleonastic pronoun) carries no semantics and is not considered an argument of the verb.



Thematic roles


Semantics also plays an important role in argument structure. Consider the following examples3 :



In (9)–(11), the verb combines with an expression that plays the role of Agent of the proposition; the Agent is essentially the “causer” or instigator of the event described by the verb. Hence, the NPs Dillon, Joelle, and Philip are all Agents of their respective verbs. This contrasts with the following examples:



In these cases, the verb combines with an expression that undergoes some change-of-state or position; we will call this the Theme of the proposition. Hence, the NPs in (12)–(14) are all Themes of their respective verbs. Thus, predicates subclassify the kinds of expressions they need into different semantic types or thematic roles. We can define thematic role in the following way:


Thematic Role: The semantic type played by an argument in relation to its predicate.


There is a limited set of thematic roles. Though linguists are not particularly concerned with details of the roles themselves, there are some major ones that bear mention; once again, let’s revisit the simple sentence:



In (15) the subject NP, the girl, plays the role of the Agent of the event, and the boy, the Theme.



In (16) the subject NP, Mitzi, plays the role of Experiencer. Roughly, the Experiencer role describes an entity experiencing some psychological or mental state.



In (17) there are three thematic roles, owing to the fact that the verb put requires three arguments. The subject argument filled by the NP the girl plays the role of Agent of the event, the direct object argument filled by the NP the boy plays the role of Theme, and the indirect object filled by the argument the closet plays the Locative role. The Locative role describes the place in which something is situated or takes place. In (18), the third argument filled by the boy plays the role of the Goal, which is roughly defined as the entity towards which something moves.



Lexical entries


Let’s assume that these properties are represented with the verb as part of its entry in the Lexicon. Consider again the verb kiss and its predicate argument structure (PAS) and thematic role features (presented in Table 6-1).



The lexical entry table shows that kiss, a verb, requires two arguments, designated by X and Y. The two arguments have particular thematic roles that need to be assigned (AGENT, THEME). Let’s assume that thematic roles are essentially features that need to be “checked off” for the sentence to be grammatical. Consider then Figure 6-5.



We have already discussed how the verb (V; kissed) merges with an NP (the boy) to form the VP (kissed the boy). We will also assume here that the V assigns a thematic role (THEME) to its NP complement, and thus we have the result appearing in Table 6-2.



Once we have assigned the thematic role to the argument position, we check-off the thematic role. As we have also discussed previously, the VP merges with the subject NP (the girl). As part of this Merge operation, the VP assigns the thematic role of Agent to the subject argument and thus we have the final result in Table 6-3.



That is, as each thematic role is assigned, the feature is checked-off in the lexical entry. When there is a match between the argument structure of the verb and the number of arguments in the sentence, including the set of thematic roles to be assigned, then the result is a well-formed sentence, with thematic roles describing “who did what to whom.” In this way, then, the lexical properties of the verb are said to project to the sentence.


Given the lexical entry for any particular verb, if there are not enough argument positions in the sentence for the thematic roles to be assigned, or if there are too many argument positions given the number of thematic roles, then the derivation of the sentence will “crash” and the result will be an ungrammatical sentence. To see what we mean, consider Table 6-4.



The verb put requires three arguments. Given the sentence:



Only two arguments are active in the sentence (the subject and object NPs, the girl and the boy, respectively). Thus, when thematic roles are assigned to these argument positions, there is one thematic role left in the lexical entry that has not been checked-off (Table 6-5).



Because there is a mismatch between the number of arguments required by the verb (see Table 6-4) and the number of arguments active in the sentence (see (19)), the features of the verb are not satisfied in the sentence, and hence the sentence is ungrammatical. Recall in the previous section that the Merge operation, by itself, is too powerful; it generates grammatical as well as ungrammatical sentences. Once we assume that the features of the verb (in this case, the argument structure and thematic role representations) must be accommodated in the sentence, the output of Merge will then be constrained to form only grammatical sentences.



Complex arguments


On some accounts, not only are NPs arguments of the verb, but so too are more complex embedded sentential clauses or Complement Phrases (CPs) (Grimshaw, 1977; Shapiro, Zurif, & Grimshaw, 1987, Shetreet, Palti, Friedmann, & Hadar, 2007). Consider the following three verbs:



As can be seen in (20)–(22), the verbs know, ask, and wonder have distinct selectional requirements; know and ask select for an NP argument while wonder does not, and all three select for a complement phrase (CP). Notice that the CP takes two different forms, one where it is headed by the complementizer that, as in (20b), while another is headed by the complementizer who, as in (20c)–(22c). These phrases are associated with complex semantic types (Grimshaw, 1977):



Those headed by a that-phrase are typically Propositions, while those headed by a wh-phrase are typically Interrogatives (there are additional semantic types, such as Exclamations and Infinitives; see Shetreet et al., 2007). Notice that these complex arguments have internal structure. Taking (23b) as an example, the argument playing the role of Proposition can be further divided into an AGENT THEME structure, where the embedded subject argument, the girl, is assigned the AGENT role and the embedded object argument, the boy, is assigned the THEME role.


Thus, a complex sentence with an embedded clause must satisfy the lexical requirements of two verbs. This can be seen in Figure 6-6.



As shown in Figure 6-6, when the embedded V (kiss) is merged with its NP complement (the boy), the THEME role is assigned (and checked-off) and a VP is formed. Moving up the tree, when the resulting VP is merged with the subject NP (the girl), the AGENT role is assigned and a CP is formed. Continuing, when the embedded clause (CP; the girl kissed the boy) is merged with the main verb (know), the PROPOSITION is assigned and a VP is formed. Finally, when the resulting main VP (knows (that) the girl kissed the boy) is merged with the main subject NP (Joe), the EXPERIENCER role is assigned and the S is formed4 ,5 .



Syntactic features of arguments


Notice that the syntactic form of an argument is not predictable from its thematic properties. For example, an embedded complement phrase (CP) can take the semantic form of a Proposition or Interrogative, as shown in (23). Furthermore, we began this exercise about argument structure showing that even simple NPs that appear to be in subject position can be assigned an Agent, a Theme, or an Experiencer role. These facts suggest that verbs select for their syntactic environments as well as thematic roles. Consider the following examples:



As shown in (24)–(27), the verb run has no complements (for now, consider a complement as the syntactic form of the argument selected by the verb); the verb push takes an NP complement; the verb give, an NP PP complement; and the verb think, a CP complement. Notice that complements do not include the subject position; both the subject and complements act as the verb’s arguments, as we discussed above when describing argument structure.


The formal name for the verb’s syntactic properties is syntactic subcategorization, also known as C-selection (Complement Selection). That is, the verb is said to subcategorize for various types of phrasal complements. Now, consider a more fully established lexical entry as in Table 6-6.



The entry shown in Table 6-6 describes the following properties of the verb discover: It has a two-place argument structure; the second (Y) argument can C-select either an NP or a CP. If there is a direct object NP argument active in the sentence, it will be assigned the Theme role; if that argument is, syntactically, an embedded clause (CP), then it will be assigned either a PROPOSITION or an INTERROGATIVE. The lexical properties of the verb will thus yield the following example sentences:



In (28a), the VP, discovered the fish, includes an NP argument playing the Theme role. In (28b), the CP argument is assigned the PROPOSITION role, and in (28c), it is assigned the INTERROGATIVE role.



Movement and copy-and-delete


A pervasive fact about languages is that they allow an element in a sentence to be displaced from one position in the sentence to another position, yet grammaticality and interpretation remain relatively stable. Consider the following pair:



It is clear that (29) and (30) are related; they both satisfy the lexical requirements of the verb kiss (it requires two arguments), and they have the same subject NP, the girl. The two sentences also have very similar objects (the boy in (29) and which boy in (30)). Thus we can say that in both cases, the Agent role is assigned to the subject position and the Theme is applied to the object position of the verb kissed. Yet, in (30) the object NP has been displaced (or moved) from its canonical position occurring after the verb (as in (29)), to a position occurring well before the verb.


Skipping many syntactic details, when an element moves, it leaves behind a movement trace (from earlier work in the 70s and 80s) or leaves behind an unpronounced copy of itself (from more recent work in the 90s). The copy must be deleted before the sentence is interpreted; otherwise the sentence will be ungrammatical and the derivation will crash (*Which boy did the girl kiss which boy?). By convention, we use <angled brackets> to signify a copy, and strikethrough to signify deletion of that copy:



Thematic roles are assigned to the base grammatical positions during the derivation of the sentence. That is, Theme is assigned to the object NP position, occupied by which boy as in (31a). Then the object NP is moved to a position earlier in the sentence, which leaves behind a copy, and the copy is subsequently deleted (31b). Because the copy and its moved counterpart are the same object, all features of the copy are shared with its moved counterpart, and hence the Theme role is transferred to the moved NP.


The type of movement characterized in (31) is called wh-movement, and it is also responsible for the derivation of clefts and relative clauses:



Like in (31), in both (32) and (33) the object NP (who) is displaced from its canonical position occurring after the verb kissed (its base position), and leaves a copy, which is subsequently deleted during the derivation of the sentence. Also similar to (31), the object receives the Theme role while residing in its base position. Unlike (31) however, in both (32) and (33) the head of the relative clause—the NP the boy—co-refers to the relative pronoun who (signified by co-indexation), and hence gets its reference (and thematic role) from the pronoun.



Argument structure, copies, and sentence processing


Given its contributions to the syntax and semantics of sentences, it should not be too surprising that verb representations have played a significant role in accounts of sentence processing. In perhaps the first attempt to examine how verbs influence sentence processing, Fodor, Garrett, and Bever (1968) found that sentences that contained verbs that accommodated two possible syntactic configurations—an NP or CP (S) complement—were more difficult to process than sentences containing verbs that accommodated only a single configuration—an NP complement. They found this to be so even though the sentences to be processed took the simplest form and were syntactically identical NP-V-NP transitive constructions. Thus, it was the verb’s potential to accommodate different syntactic structures (i.e., their implicit lexical representations), and not the surface realization of one or the other of these structures, that appeared to contribute to sentence processing performance. Fodor et al. (1968) used paraphrase and anagram tasks to discover the relation between the complexity of verb representations and sentence processing performance. Similar findings were reported by Holmes & Forster (1972) using rapid serial visual presentation (RSVP) and by Chodorow (1979) using time-compressed speech.


In a related series of experiments, Shapiro and colleagues (e.g., Shapiro, Zurif, & Grimshaw, 1987, 1989; Shapiro, Brookins, Gordon, & Nagel, 1991) discovered the relation between the number of argument structure configurations and sentence processing complexity, using a cross-modal lexical interference task (Box 6-1). Briefly, verbs accommodating different numbers of argument structures were inserted in sentences with similar, simple, surface forms. These sentences were presented to normal listeners, who had to complete a secondary task that was presented in the immediate temporal vicinity of the verb. Verbs that entailed more argument structure possibilities yielded greater processing load relative to verbs that entailed fewer possibilities, suggesting that once the verb is encountered in a sentence, all of its possible argument structure arrangements are activated. One reason for such exhaustive activation is that it allows for on-line thematic role assignment, as we discussed earlier. That is, once the verb is encountered and activated, so too are its argument structure and thematic roles, setting the stage for further operations of the sentence processor (see, for example, Clifton, Speer, & Abney, 1991; Pritchett, 1988; Boland, Tanenhaus, & Garnsey, 1990).



Box 6-1   Experimental Techniques


Offline questionnaires


Participants are given all the time they need to indicate their judgments, ratings, classifications, sentence completion, or selection of alternative interpretations by writing.







Also, our foray into the syntax of movement has important implications for sentence processing. As we shall show shortly, when a listener who is attempting to understand a sentence encounters a direct object position that is “empty”—where the direct object NP has been displaced to a position that occurs before the verb—the listener appears to activate that NP, even though it is not heard or seen at the post-verb position. This is a remarkable finding, and suggests that linguistic theory does have something to offer those of us who are interested in how the brain comprehends language. Furthermore, constructions with movement turn out to be particularly problematic for some individuals with aphasia, as Chapter 10 reveals. With these linguistic preliminaries out of the way, we now turn to sentence processing.



Comprehension


How do we go from a series of sounds or visual letters, identify words and phrases, and extract complex meaning from them? At what point in time during the comprehension process are different types of information (syntactic, probabilistic, world knowledge, etc.) taken into account when constructing an interpretation? These are central questions in the study of language comprehension. First, we will discuss how words are organized in and retrieved from our mental lexicon. Next, we will discuss how and when information can guide the interpretation of a sentence.



What’s a word?


A word or lexical entry is a verbal label for a concept. Many types of information are bundled under this label: the word’s written or phonological form (e.g.: dog or /d⊃g/), its syntactic category (noun, verb, preposition), its number (singular or plural), its gender (feminine, masculine, neutral), its argument structure, and of course, the concept it refers to (a four-legged animal that barks and fetches objects), and others.


Psychologically, all this information, for each word, has to be easily retained and retrieved from memory. How a word’s information is encoded is referred to as the mental representation of the word, while the access of a word’s mental representation is referred to as lexical retrieval or access. While the types of information composing a word are largely agreed upon, there is little consensus on the nature of a word’s mental representation.


Possibly, a word’s representation is the psychological counterpart of a dictionary entry: under its form (written or phonological), a word’s detailed information could be automatically retrieved as a block. A word would be a unit that is used as a building block by the comprehension system. This view tends to be assumed by modularist, form-based accounts of sentence comprehension, which we will discuss below.


Another possibility is that words are not mentally stored as wholes, but rather as groups of properties. That is, each type of information would be encoded independently of the word and it would be the combination of specific attributes (e.g., a noun, written form “dog,” “singular,” “four-legged animal,” “that barks,” etc.) that would come together to form the word entry “dog.” This view tends to be associated with connectionist, constraint-based accounts of sentence comprehension, though it is by no means accurate to suggest that only connectionist accounts can accommodate bundles of properties.


These two views of word representation also differ on the nature of syntactic constraints, the rules governing the combination of words. For form-based accounts, these rules exist independently from word representations. For constraint-based accounts, syntactic rules do not exist on their own; rather, information on how a word can be used in a sentence is integrated into each word’s lexical entry. The special status attributed to independent syntactic rules forms a cornerstone of form-based models of sentence processing, as we will see later.

Stay updated, free articles. Join our Telegram channel

Jan 6, 2017 | Posted by in PSYCHOLOGY | Comments Off on Linguistic and psycholinguistic foundations

Full access? Get Clinical Tree

Get Clinical Tree app for offline access